OPBSResearch

Must Know

An RCT is a prospective analytical study comparing at least two treatment groups formed by randomisation.
A well-run RCT minimises bias and is level 1 evidence on the Centre for Evidence-Based Medicine scale.
Being labelled an RCT does not guarantee high quality.
Randomisation reduces selection bias and balances known and unknown confounders.
Balancing unknown confounders is the main advantage over matched nonrandomised prospective studies.
Allocation concealment hides the next allocation until assignment. It protects against selection bias.
Blinding keeps treatment allocation hidden after assignment and for the whole trial.
Surgeons cannot be blinded in a surgical trial. Assessors and sometimes patients can.
Strict inclusion criteria raise internal validity but reduce external validity and slow recruitment.
The primary outcome is set before the study and drives the sample size.
Many secondary outcomes raise the risk of false-positive results.
Trials should be registered publicly before recruitment, for example on clinicaltrials.gov.

Trial Designs

Design	Feature	Pros and cons
Parallel	Each subject receives one intervention. Commonest design	Simple. Example Moseley 2002 placebo arthroscopy
Crossover	Each subject gets every treatment in sequence and acts as own control	Removes confounders, smaller sample. Longer, carry-over effects, needs washout. Rarely feasible in surgery
Factorial	Two or more interventions tested at once, for example 2 by 2	Only design that tests interactions between treatments. Example FLOW trial

A crossover design differs from treatment crossover, which is patients switching arms in any trial.

Randomisation Methods

Method	How	Limitation
Simple	Coin toss or random number	Group sizes may be unequal
Block	Randomise within blocks sized as multiples of the number of groups	Best for small samples. Baseline covariates may still differ
Stratified	Blocks formed by a covariate such as sex, then randomised	Tedious with many covariates. Covariates must be known in advance. Unknown ones unbalanced
Adaptive	Each new subject assigned using their covariates and previous allocations	Complex

Blinding Levels

Level	Who is blinded
Single	Subjects
Double	Subjects and investigators
Triple	Subjects, investigators and outcome assessors or analysts

Sample Size and Effect Size

Inputs are power, type 1 error, expected effect size and variability in the control group.
A highly variable outcome needs a larger sample. A large frequent effect needs a smaller one.
Oversized trials waste resources and expose extra patients to inferior treatment.
The MCID is the smallest change patients find clinically relevant.
Superiority, equivalence and noninferiority trials each use different sample size formulae.
Only about a third of 215 high-impact RCTs reported sample size calculations adequately.

Conduct and Analysis

Measure height and weight. Do not rely on self-report.
Assessor-measured function suffers interobserver variability. Establish reliability first and blind assessors.
PROMs avoid assessor bias but culture can change how patients interpret them.
Loss to follow-up rises with time. Above 20 percent introduces a higher level of bias.
Interim analysis by a Data Safety and Monitoring Board can stop a trial for early superiority or harm.
Other reasons to stop are futility, redundancy and loss of funding.
ITT analyses all patients in their allocated arm whatever they received.
PP analyses only fully compliant patients and is a subset of the ITT population.
ITT preserves sample size and power and gives conservative, diluted effect estimates.
PP alone carries more bias and can hide complications or deaths.
FDA and ICH E9 recommend routine ITT. CPMP recommends ITT as primary and PP as supportive.
Agreement between ITT and PP makes a result more likely to be true.

CONSORT

CONSORT is the reporting standard for RCTs and a template for appraising them.
First statement 1996, revised 2001 and 2010.
CONSORT 2010 has a 25-item checklist and a flow diagram.
CONSORT-PRO (2013) covers trials with patient-reported outcomes.
Binary outcomes should report both absolute and relative effect sizes.

Exam Traps

Allocation concealment is not blinding. Concealment ends at assignment. Blinding continues afterwards.
Stratification does not balance unknown covariates.
No standard minimum acceptable loss to follow-up exists.
ITT includes patients who drop out before starting treatment.

Related pages. Level of Evidence, Grades of Recommendation

Recent Review. Noninferiority Studies

A nonsignificant P value in a superiority trial means insufficient evidence of a difference. It does not show the treatments are the same.
With β set at 0.1 to 0.2, a superiority trial carries a 10 to 20 percent chance of missing a true difference.
Very large or overpowered datasets yield small P values for differences too small to matter clinically.
Noninferiority asks whether a new treatment is no worse than control within a set margin. Equivalence asks whether it is the same within a two sided margin.
Noninferiority uses a one sided test, so it often needs a smaller sample than a two sided equivalence test.
Noninferiority is shown when the lower bound of the one sided 95 percent CI of the difference sits above the negative margin.
Equivalence is shown when the two sided 90 percent CI of the difference lies wholly inside both margins.
The noninferiority margin must be fixed a priori as the smallest clinically meaningful between group difference.
The MCID is often borrowed as the margin, yet it measures change within one patient over time.
The margin should never be smaller than the minimal detectable change of the outcome instrument.

Recent Review. The Fragility Index

Fragility index is the fewest patients whose outcome must flip for a significant trial result to lose significance.
It applies only to binary outcomes such as death or revision. It is not used for continuous or time to event outcomes.
Calculate it by moving one non-event to an event in the arm with fewer events, then rerunning Fisher exact test until P exceeds 0.05.
A lower fragility index means a more fragile result. An index of zero can follow use of the wrong statistical test.
Reverse fragility index is the fewest outcome changes needed to turn a nonsignificant result significant.
Fragility quotient is the fragility index divided by sample size, allowing comparison across trials of different size.
No threshold value marks a trial as fragile or secure.
The fragility index correlates closely with the P value and rises with sample size. It does not measure size of treatment effect.
Apply it only to a binary primary outcome. Secondary outcomes are underpowered, so they look fragile by design.

Recent Review. The Table I Fallacy

Table I of an RCT lists baseline characteristics so readers can look for imbalance in prognostic variables.
Randomisation aims to balance known and unknown prognostic variables but does not guarantee it.
Chance imbalance is more likely in small pilot trials and cluster randomised trials.
Significance testing of baseline variables after proper randomisation is called the Table I fallacy.
CONSORT item 15 advises against statistical tests of baseline characteristics.
Any baseline difference after randomisation is by definition due to chance, so testing it is illogical.
With α at 0.05, about 1 in 20 baseline variables will show P below 0.05 by chance alone.
A nonsignificant baseline difference in a strong prognostic variable can still confound the result.
Covariates for adjustment should be chosen a priori in the protocol on clinical grounds as known predictors of outcome.

Recent Review. Adaptive and Platform Trial Designs

An adaptive trial allows prespecified changes to the protocol while data are still accruing.
Adaptation rules must be fixed before the first patient is enrolled, often tested by statistical simulation beforehand.
Dose-response drug trials shift recruitment towards better performing doses at a prespecified interim look.
Pragmatic adaptive trials usually adapt by re-estimating sample size or applying futility stopping rules (START REACTS is an orthopaedic example).
A platform trial compares several simultaneous interventions against one shared control arm in a single condition.
An umbrella trial tests one intervention across different patient groups, often defined by a molecular biomarker.
RECOVERY (COVID-19) is the best known platform trial and identified dexamethasone as the first effective treatment.
Platform downsides include amendment knock-on delays, early stopping that may miss real benefit, and interaction effects between comparisons.
An interaction effect exists when the effect of one variable on outcome depends on the level of a second variable.

Recent Review. Fragility Index and P Values

The fragility index (Walsh) is the number of event changes needed to turn a significant result nonsignificant.
The reverse fragility index is the number of event changes needed to make a nonsignificant result significant.
Alpha is the accepted false positive (type I error) rate, conventionally set at 0.05.
A power analysis sets the minimum sample size to detect the minimal clinically meaningful effect.
Trials sized to the power calculation will often finish with p values near 0.05 and a low fragility index.
The fragility index correlates almost perfectly with the p value (R above 0.9) and adds no new information.
Enlarging a trial lowers the p value and raises the fragility index while the effect size stays unchanged.
Post hoc (observed) power is a one to one function of the p value and adds no information.
CONSORT reporting already gives central tendency, variance (SD or 95 percent CI) and the p value.

Source

Lin HA, Henry PDG, Wasserstein D. Randomized Controlled Trials. In. Aaron RK, editor. Orthopaedic Basic Science. Foundations of Clinical Practice. 5th ed. Philadelphia, Wolters Kluwer, 2021.

Practise Type X questions on this topic

Reviewed by Professor Phong Tran, Head of Orthopaedic Surgery, Western Health. Last updated 17 October 2026.